NeuroCognica
The research
The products are what the research produced. This is the work they came out of — and the parts still unresolved.
Read this as work in progress, not findings. None of it is peer-reviewed, none of it is published in a journal, and nothing here should be treated as an established result. It is one person's ongoing investigation, stated openly so it can be argued with.
Three questions
1. Can a machine reflect a person back honestly?
Personality instruments mostly hand you a label and treat it as a fact about you. The interesting question is different: can a system organise what someone says about themselves, show its evidence, admit what it cannot support, and be wrong in ways the person can see and challenge?
That means a fixed taxonomy rather than a model's free improvisation, deterministic assignment rather than an impression, and a visible completeness figure rather than a confident blank. Mirrorborn is that question made testable.
Unresolved: the hardest station to elicit honestly is the one a person has disowned. In current testing it frequently comes back empty. That is a real limitation, published rather than smoothed over.
2. What would a constitution for a machine have to say?
"Alignment" usually means a model trained to decline certain prompts. That is a behaviour, not a constitution — it has no articles, no appeal, no record, and no way to tell whether it was followed.
The working document behind these systems is The Charter of Cognitive Sovereignty: named laws rather than vibes, articles that can be cited, and a requirement that a refusal be attributable and contestable. Its principles are what the Sentinel implements — separating who may act from what may be made from how it executes, and sealing every decision so the record survives the person who made it.
Unresolved: a constitution enforced only by the vendor who wrote it is a policy. What it would take for such a thing to bind its author is genuinely open.
3. What does sovereignty actually cost?
Local-first is easy to say and expensive to mean. Running capable models on one person's machine forces real constraints: quantisation, model size, cold starts, and the plain fact that a laptop with no GPU is a different order of magnitude, not a slower version of the same thing.
Every measurement we take on that gets published, including the unflattering ones. A CPU-only render taking close to an hour is stated on the product page before purchase, not discovered afterwards.
Why this is funded by selling software
There is no grant and no investor. ChronoSophia pays for the research, which is why it has to be a real product rather than a demo — and why the research standards and the product standards ended up being the same standards.
It also means the work answers to nobody's roadmap. That is the compensation for it being slow. If you want to help without buying the product, back the work directly.
If you want to check the work
The claim that distinguishes this from ordinary marketing is that the evidence is published, including where it fails:
- The Sentinel Security Protocol — measured refusal and false-refusal rates, the methodology, and the findings still open
- The raw per-probe results — the data behind those numbers, not just the summary
- The capability accounting — what is proven, what is wired but unwitnessed, and what is absent
- The Shaolin Virtue Compass — an interactive learning map for virtues, ethics, laws, morals and principles
Participating
When Mirrorborn opens, taking the assessment is optional and contributing your answers to the research is a separate choice you can decline without losing anything. Why that separation is not negotiable is explained in what we ask, and why.